Sign in to view source links and access this dataset
Description
The Exclusively Dark (ExDARK) dataset contains images from 10 different low-light conditions, ranging from very low-light to twilight. It includes annotations for 12 object classes, such as Dog, People, and Car, at both the image and object bounding box levels. The dataset was uploaded to Hugging Face by SatwikKambham in October 2023.
Use Cases
Train object detection models based on the 12 annotated object classes like people, cars, and bicycles.
Benchmark model robustness across 10 different low-light conditions described in the dataset.
Develop image classification systems for low-light scenes using the provided image-level labels.
Strengths
Annotations are provided at two levels: image class and local object bounding boxes.
Covers 10 distinct low-light conditions, from very low-light to twilight.
Includes 12 object classes, which are similar to the standard PASCAL VOC benchmark.
Limitations
Description metadata is limited; actual data quality requires manual inspection after download.
Column-level documentation is absent; field semantics must be inferred after download.
Row count is unknown, which may limit suitability assessment.